INFOCOMMUNICATIONS Linearity Improvement of GaN HEMT for Wireless Communication Applications Kazutaka INOUE*, Hiroshi YAMAMOTO, Ken NAKATA, Fumio YAMADA, Takashi YAMAMOTO and Seigo SANO For GaN HEMTs to be widely used for microwave amplifiers such as point-to-point backhaul systems, good linearity is required. This paper describes our recent achievement in improving linearity by using a newly constructed large signal model of a 0.4 µm GaN HEMT. The model analysis revealed that the intermodulation distortion (IMD) at a backed-off region of more than 10 dB is determined by the sub-threshold gm profile; in other words, steep rising gm profile degrades IMD. We created a GaN HEMT that has a thin n layer inserted in the buffer (ini-buffer) structure, and achieved a significant IMD improvement in the backed-off region of more than 8 dB. Keywords: GaN HEMT, linearity, gm, large signal model, microwave 1. Introduction The modern digital telecom systems require the excel- lent linearity to the amplification devices, to realize the Gallium Nitride (GaN) is desirable material for the ap- high bit-rate data transformation. In the case of the BTS plication of high-power and high-speed operation electron systems, the distortion cancelling techniques, such as the devices because of its excellent properties such as large digital-pre-distortion (DPD), have been widely adopted to energy band gap and high saturated electron velocity. We suppress the non-linearity of the amplification device. The have released the GaN high electron mobility transistors situation of the P-to-P backhaul amplifier is quite different. (HEMTs) on SiC substrate, mainly targeting for the ampli- The higher frequency amplification makes it difficult to fiers of the base station transmitter systems (BTSs) at the adopt the linearization techniques. In addition, the system frequency range of below 3 GHz. GaN HEMTs have be- does not allow increasing complexity of the amplifier sys- come widely employed to the modern BTS amplifiers such tems. Therefore, pre-distortion techniques are not effective as long-term evolution (LTE), which requires the high out- or applicable for P-to-P systems, or in other words, the put power with excellent efficiency. amplifier device for P-to-P is required high linearity in itself. Figure 1 shows a schematic illustration of the present This paper describes our study for the linearity improve- communication infrastructure network. Among them, ment of the GaN HEMT. point-to-point (P-to-P) backhaul network systems require the high-power and high-efficiency devices, although GaAs devices have been used in these communication systems. GaN HEMT is suitable for such high-frequency and high- 2. Linearity of GaN HEMT power applications including P-to-P systems, thus many ef- forts have been dedicated to the development. The intermodulation distortion (IMD) is one of the useful indexes to estimate the linearity. In the IMD evalu- ation, 2 tone input signals of the frequency of f1 and f2 are used and the input vs. output property is measured. In the ideal case, the frequency components of only f1 and f2 are observed (Fig. 2 (a)). While in the actual amplification, the frequency components besides f1 or f2 are found in the out- put signals (Fig. 2 (b)), which are the IMD signal compo- nents. For example, the 3rd order compornents of IMD signals (IM3) emerge at the frequencies of 2f1-f2 and 2f2-f1 (a) (b) Fig. 1. Present communication infrastructure network Fig. 2. (a) Ideal amplifier (b) Actual amplifier 48 · Linearity Improvement of GaN HEMT for Wireless Communication Applications as shown in Fig. 2 (b), and these power levels correspond to the degree of the non-linearity. This study has been done, based on the GaN HEMT structure as shown in Fig. 3 (a). Actually, the AlGaN/GaN epitaxial structure was grown on the 4 inch semi-insulating SiC substrate by utilizing Metal Organic Chemical Vapor Deposition (MOCVD). The gate length was set to 0.4 µm and the AlGaN barrier layer thickness was tuned by refer- ring the maximum drain current (Imax) versus the pinch off property of the channel. (a) (b) Fig. 4. Small-signal equivalent circuit model The outline of the model construction is as follows. Firstly, the multi-bias S-parameters were measured at vari- ous bias conditions, and then small-signal equivalent circuit parameters were extracted. Figure 4 shows the small-signal equivalent circuit for the analysis. Figure 5 shows the ex- tracted result of gm as a function of the extrinsic gate volt- Fig. 3. (a) Conventional GaN HEMT (b) IMD property age (Vg) and the extrinsic drain voltage (Vd). The unit gate width (Wgu) of the measured pattern was 50 µm and the total gate width (Wgt) was 300 µm. For a large-signal mod- eling, gm, Cgs, Cgd, and Rds were treated as non-linear pa- rameters, while the others were as the linear ones. GaN exhibits two times higher saturated electron ve- The DC modeling was also performed by treating gm locity and eight times larger critical breakdown field than and Rds as non-linear elements. In modeling a GaN HEMT, those of GaAs. IM3 of below -40 dBc is supposed to be re- taking into consideration of the current collapse is one of quired to the P-to-P systems, and the class-A operation is ideal to realize such excellent IM3 performance. As already mentioned, GaN HEMT realizes higher output power den- sity by utilizing the higher voltage operation. Actually, the typical operation voltage of GaN HEMT is 24 V, and real- izes about ten times larger power density than that of GaAs FET. The higher voltage operation also forces GaN HEMT deep class-AB operation to keep equivalent DC power con- sumption to 10 V biased and class-A operation GaAs FET. Figure 3 (b) shows the typical IM3 profiles of the GaN HEMT of 24 V and class-AB operation, and GaAs FET of 10 V and class-A operation. The IM3 profile of the GaN HEMT exhibits the plateau profile, while the GaAs FET shows the plateau-less profile. The plateau-less profile of class-AB GaN HEMT is essential to replace the GaAs FET Fig. 5. Extracted gm vs. extrinsic Vg, Vd in the P-to-P application. Thus, we set the goal to realize the plateau-less profile and started with the GaN HEMT model construction and the operation analysis. (a) (b) 3. Model Construction and Analysis Many models have been proposed and utilized for the large signal analysis of microwave transistors. Among them, we focused on Angelov model(4), (5) and have constructed the model for our 0.4 µm GaN HEMT. Angelov model fea- tures the straightforward expression utilizing the equiva- lent circuit parameters, and relatively short computing time to complete the analysis. Fig. 6. (a) IV without voltage stress (b) IV with voltage stress SEI TECHNICAL REVIEW · NUMBER 78 · APRIL 2014 · 49 the key issues to establish its accuracy. The measurements The time-domain gm plot of rather small signal case were done by the µ-second order of pulsed signals, and the (Pin = -10 dBm) exhibits approximately sinusoidal shape. quiescent drain bias was set to 24 V, which is the typical In the case of the input power level of 0 dBm, the maxi- drain voltage in our targeted operation. The drain current mum region is clipped at first. And then, the plot of the was modeled as the following Angelov formula: input power of +10 dBm that shows the minimum side of the gm profile is also clipped. The maximum, average and (1) minimum values of the analyzed gm profiles were plotted in Fig. 9. Through the careful comparison of Fig. 8 and Fig. 9, the on-state clipping power level is around -5 dBm, where IPK0, Ψ, α, LAMBDA, LSB0, and VTR are the and the off-state clipping is around +2 dBm. fitting parameters as defined in (5). The harmonic-balance analysis at the drain bias of 24 V, the drain current of 10% of Imax and the frequency of 10 GHz has been done using the obtained parameters. Figure 7 (a) shows the simulated 2-tone Pin-Pout property, and Fig. 7 (b) shows the IM3 profiles. The obtained results are in accordance with the measured profiles up to the input power level of 10 dBm. Through this comparison, we concluded the constructed model is appropriate for the in- tended linearity analysis. (a) (b) Fig. 9. Analyzed gm plot versus input power Figure 10 shows the obtained gm profile through the measurements and the analysis. The solid circle (●) shows the quiescent bias point and the RF swing occurs in accor- dance with the input power levels. At first, we assumed the complex on-state gm warping Fig. 7. (a) Pin-Pout profile (b) IM3 profile corresponds to the inferior IM3 property of the GaN HEMT. Therefore, we set the rectangle gm profile as shown in Fig. 10 and analyzed the IM3 profile. On the contrary to the assumption, the analyzed IM3 profile of the rectangle Then, the analysis to improve the linearity of the GaN gm case rather degraded around the 15 dB backed off re- HEMT of class-AB operation was performed. It is com- gion as shown in Fig. 11. The rectangle gm enhances the monly recognized that the non-linearity originates in the plateau shape of the IM3 profile. In addition, it is revealed amplitude or the phase, and the amplitude non-linearity is through the analysis that the marked region in Fig. 11 cor- strongly affected by the gm profile. Thus, our analysis is fo- responds to the abrupt gm rising region from off- to on-state cused on the gm property at first.